Home › Learn › Lesson 34

Fibonacci extension — setting the target after the pullback ends

By TradingPrimer · Published 5 Sep 2026

Most guides on Fibonacci extensions list the levels, draw an arrow at 161.8% and stop. Two things go missing. The first is that our own course material pulls in two directions on whether this tool should be used to set targets at all — the slide says find the target, the course note on Fibonacci says read the force and stop there — and that disagreement turns out to be the most useful thing in the lesson once you see what each one is actually talking about. The second is that nobody prices the ladder. Do it, and the five rungs stop looking like a menu: one of the four steps carries more than half of everything the ladder has to give, and the moment price is actually moving, the arithmetic that made the far rung attractive quietly reverses.

Title card. Every measured claim in this lesson lives on the two charts further down, which are drawn from the prices themselves rather than sketched.

What does a Fibonacci extension actually measure?

How far the next leg might run — expressed as a multiple of the leg that already ran. That is a different job from the one Lesson 33 gave the retracement tool, and it uses a different measurement.

A retracement asks a question about the past: how much of the impulse did the pullback take back? Both of its anchors sit on a move that has already finished, so the answer is a fact. An extension asks a question about a leg that has not happened yet. It cannot be a fact, and this lesson is largely about what it is instead.

Our course slide states the job in one line: finding the target of an impulse wave once the correction or bounce has finished. Three words in that sentence are the whole precondition — once, has, finished — and Section 2 is about why they are not just style.

The ladder our course teaches has exactly five rungs: 100%, 127.2%, 138.2%, 150%, 161.8%. There is no 261.8% on it, and we have not added one from elsewhere.

Look at what those five numbers are made of, because almost nobody points this out and everything below follows from it. Three of them are the retracement ladder stacked on top of a complete leg: 138.2 = 100 + 38.2, 150 = 100 + 50, 161.8 = 100 + 61.8. The odd one out is 127.2, which is √1.618. So the extension tool is not a new set of numbers at all — it is the same ratios, measured from the other end.

That construction leaves the ladder uneven. The gaps between consecutive rungs are 27.2, 11.0, 11.8 and 11.8 points of impulse. The first gap is 2.36 times the average of the other three. Hold on to that, because Section 5 turns it into money.

Where do the three anchors go, and why must the correction finish first?

A at the start of the impulse, B at its end, C at the end of the correction. Each rung then sits at C + k × H, where H is the impulse height and k is the rung. In a falling market every sign flips and nothing else changes.

The worked example this whole lesson runs on, stated once so every number below can be checked by hand: an impulse from A = $28,000 to B = $34,000, so H = $6,000. The correction retraces half of it and ends at C = $31,000. Entry at C, stop at A — below A the structure that produced all of this is simply gone — so risk = $3,000.

The two teal bars in that chart are the definition of the 100% rung, drawn rather than asserted. One measures the impulse. The other stands on the end of the correction and reaches up to $37,000. They are the same length because the rung is the impulse, moved.

Now the precondition. Anchors A and B sit on a move that is over; you can be wrong about which move matters, but not about whether it happened. Anchor C is different. Until the correction has actually turned, C does not exist — and since every rung is measured from C, none of the five numbers exists either. A ladder drawn on a correction still in progress is not an early reading. It is five prices generated from a coordinate you invented.

This is not our own rule. Our course note on the Fibonacci tool insists on both halves before the tool comes out at all: a completed impulse and a correction that has formed and turned back. The slide says the same thing in five words: once the correction has finished.

There is a practical tell. If the level you are quoting moves every time a new candle closes, you are not measuring — you are following. A finished correction gives you a C that stops moving, and rungs that stop moving with it.

Is a target the same thing as a prediction?

No, and the distinction is the reason this lesson can exist at all. Our own course material pulls in two directions here, and the honest thing is to say so rather than quietly pick a side.

The slide teaches extension as a way of finding the target. The course note on the Fibonacci tool, which is the more senior of the two sources, says the opposite in as many words: read the depth of the correction, conclude whether force survived, and stop there — it files "the next wave will reach X" under mistakes, alongside mis-measuring the wave.

Both are right, because they are talking about two different sentences.

Everything priced below is the second sentence only. Not one number here says how often price arrives at a rung, because we have not measured that and neither has anyone who quotes a figure at you without saying how they counted. What can be priced is the arithmetic of the decision — and that turns out to say a lot more than most people expect.

What does each rung actually pay?

Reward at rung k is k × H, and nothing else. Entry is at C, the target is at C + k H, and C cancels. So the reward does not depend on how deep the correction went, or on the price of the asset, or on where the stop sits — only on which rung you picked.

Put that over the $3,000 of risk and the ladder becomes a table. Break-even win rate throughout is 1 ÷ (R + 1) — the win rate the trade would need in order to be worth taking, and nothing more than that.

Read the table on its own and it says something simple and slightly reckless: reaching further is free. Risk was fixed the moment you chose an entry and a stop, so every extra rung is reward at no extra cost, and the 161.8% column looks like the obvious answer.

It is not, for two reasons that take the rest of this lesson. The first is that the ladder does not hand out its benefit evenly. The second is that "free" stops being true the moment price actually starts paying you.

Why does the first rung buy more than half of everything?

Because the ladder is lopsided, and the currency it pays in — break-even win rate — is not linear in the target.

Across the whole ladder, from 100% to 161.8%, the break-even requirement falls by 9.72 points, from 33.33% to 23.61%. That is the entire budget. Here is how the four steps split it.

The step from 100% to 127.2% takes 5.12 of the 9.72 points — 52.6% of the whole ladder, bought in one move. The three steps after it share 1.65, 1.57 and 1.39 between them.

This is not an artefact of the example. Rerun it at other correction depths and the share barely moves: 52.1% at a 38.2% correction, 52.6% at 50%, 53.2% at 61.8%, 54.3% at 78.6%. The reason it is so stable is that it is a fact about the rungs, not about the trade. A 27.2-point gap followed by three gaps of about 11 will always behave this way.

So the practical version is short. If you are going to reach past a measured move at all, 127.2% is where nearly all the value of reaching is. Everything beyond it is a real improvement, but a small one — and Section 6 shows it is not free after all.

Price has reached 100%. Bank it, or run?

This is a different question from the one Section 4 answered, and it has close to the opposite answer. Before entry you were comparing target choices at a fixed risk, so further was free. Now price has paid you, and running on means putting a real gain back on the table to reach for a bigger one.

Our course note is strict about the only stop adjustment allowed once a trade is working: you may move the stop to break-even, and you may not move it to a Fibonacci level. So there are exactly two honest versions of "run on", and both can be priced.

With the stop moved to break-even. Banking at rung k₁ gives R₁ every time. Running to k₂ gives R₂ with probability q and nothing otherwise. Setting the two equal gives q = R₁ ÷ R₂ — and since both R's are the same rung multiple over the same risk, everything cancels:

Look at the middle column. 78.6%, 72.4%, 66.7%, 61.8% — that is the retracement ladder read backwards, and it is exact rather than approximate. 61.8% is 1 ÷ 1.618; 78.6% is 1 ÷ 1.272, which is 1 ÷ √1.618. Because the extension rungs are generated from the golden ratio, the ratios between them fall back onto the numbers you already know from retracements.

Now the part that inverts Section 5. The nearest rung is the most expensive one to reach for. Running from 100% to 127.2% demands the highest continuation rate on the table — 78.6% — because you are risking a gain you already hold in exchange for a small increment. Running all the way to 161.8% demands the lowest, 61.8%, because the prize is worth the exposure.

So 127.2% is the cheapest rung to plan and the dearest to chase, and both readings are correct. Nothing has changed except when you are standing. Before entry, risk is sunk and reward is free. Mid-trade, the gain is the thing at stake. Any advice that gives you one answer without asking which moment you are in is answering a question you did not ask.

If you find yourself unable to choose, that is the correct reaction rather than a failure of nerve, and there is a standard resolution — taking part of the position off at the near rung and letting the rest run to the far one. It is a lesson of its own later in the path, and doing it properly needs sizing rules this lesson has not set out, so we will leave it named rather than half-taught.

What if you anchor from A instead of C?

Every rung lands lower by exactly the depth of the correction. In this example that is $3,000 at every single rung, which — because the stop sits at A — is exactly 1.00R.

The constant is not a coincidence. The two conventions differ by C − A, which is the depth of the correction, which is also the entry-to-stop distance. Change the rung and the gap does not move.

Now find the trap in that table. The two-point 150% and the three-anchor 100% are both $37,000. Two traders using different conventions can quote the same price at each other, agree, and be describing different levels — and neither has any way to notice. That is the argument for stating your convention in your notes rather than assuming there is only one.

What if the rung and a real level disagree?

Take the nearer of the two. Our course is explicit that Fibonacci is never used on its own, and the reason bites harder on exits than on entries.

Lesson 33 established the rule for entries: the number is not the level, the level is the level. A price holds because orders sat there, not because a ratio pointed at it. Exits inherit that logic exactly. A rung is an arithmetic consequence of where you put three anchors. An old high or a resistance zone is a place where sellers actually appeared.

Put a number on the choice. Suppose there is an old swing high at $39,000 and your 138.2% rung is at $39,292. Deferring to the level costs $292 — on $3,000 of risk, that is 0.10R. A tenth of a unit of risk to exit where other people's orders are rather than 292 dollars beyond them.

The alternative that always tempts — nudging the anchors until the rung lands on the level — is worse than either. It preserves the appearance of a measurement while removing the measurement, and afterwards you cannot tell which of your levels you found and which you arranged.

Does any of this change in a falling market?

No. The construction is symmetric and our course carries both cases side by side. A falling impulse from $34,000 down to $28,000, with a bounce back to $31,000, gives rungs at $25,000 (100%), $23,368 (127.2%), $22,708 (138.2%), $22,000 (150%) and $21,292 (161.8%). The R:R column, the break-even column and every continuation rate in Section 6 come out identical, because none of them depends on direction.

When is everything above wrong?

Four conditions, and the first two void the tool rather than adjust it.

• The market is not trending. Our course opens this section with the market has to be trending, and the note is firm that in a sideways range the tool loses its measuring power altogether. In a range there is no impulse and no correction — there are two sides of a box, and labelling one of them A–B does not make it a wave. Diagnose the state first; Lesson 20 is that diagnosis.
• The correction went deep. Lesson 33 reads a pullback into the 61.8–78.6% band as force largely spent. The arithmetic here still computes — it will compute for any inputs at all — but it is pricing a next leg that the evidence says may not be coming. The table is only as good as the premise it sits on.
• The instrument does not travel that far. The ladder is scale-free and will happily print $40,708 on something that has never managed a move of that size on your timeframe. Check the rung against the largest run the thing you are trading actually makes.
• The costs are missing. Every break-even figure here assumes one exit, no fees and no slippage. All three push break-even up rather than down, and they push it up most on the near rungs, where the reward is smallest relative to the friction.

What are the most common mistakes here?

What else do people ask about Fibonacci extensions?

PRACTICE CORNER

Do this once on your own charts, because it settles Section 6 with your own numbers instead of ours. Find a completed impulse with a finished correction, drop the three anchors, and write down all five rungs in prices. Then write two lines next to them: which rung is the plan, and what you will do if price reaches 100% and hesitates. Answer that second line before price gets there — the whole point of this lesson is that the mid-trade answer is not the pre-entry answer, and that the difference is 78.6% versus 61.8% rather than a matter of nerve. Finally, mark where the nearest old high sits relative to your chosen rung, and note which one you would actually exit on.

You need a platform whose extension tool takes three anchors rather than two, and a chart you can read an old swing high off at the same time. These are the three exchanges this site uses for its own worked examples; all three have the tool, and the exercise costs nothing.

We may earn a commission if you open an account through these links, at no cost to you. It does not change what is written above.

Binance→ OKX→ Bybit→

Educational content only — not financial advice, and not a trade recommendation. Every figure on this page comes from one worked model and each can be reproduced by hand from the numbers given: a rising impulse from $28,000 to $34,000, so H = $6,000; a correction that ends at $31,000, which is half of it; entry idealised at that low and the stop at $28,000, so risk = $3,000. Extension rungs are C + k × H, reward at rung k is therefore k × H, and break-even win rate is 1 ÷ (R + 1) throughout. Entering exactly at the correction low is a simplification — in practice entry sits a little above it, which lowers every R in the same proportion and does not change the ranking of the rungs. The continuation rates in Section 6 come from setting the expected value of running on equal to the value of banking: q = k₁ ÷ k₂ with the stop at break-even, and q = (R₁ + 1) ÷ (R₂ + 1) with the original stop in place. The 52.6% share was rechecked at correction depths of 38.2%, 50%, 61.8% and 78.6%, giving 52.1%, 52.6%, 53.2% and 54.3%. Every percentage on this page that describes a trade’s odds is a threshold — the win rate or continuation rate a decision would need in order to be worth making. No historical hit rate is quoted anywhere, because none was measured, and we are not aware of a published one we would trust. The five-rung ladder, the instruction to anchor after the correction has finished, the requirement that the market be trending and the warning never to use Fibonacci on its own are all taken from the polarity and Fibonacci section of the TradingPrimer slide course; the distinction between reading force and predicting a destination comes from the course notes on the same tool. Sources: that course material, plus the arithmetic above.